
ABSTRACT
The signal processing is a research object in the most electronics laboratories,
often since for their first years of existence. Work is intensified with the appearance of the
digital signal. The denoising is an essential method in signal processing. The implementation
of the discrete wavelet transforms (DWT), the choice of the thresholds and the functions of
thresholding were peeled and studied in the purpose of satisfying the constraints of the
targeted application. These constraints are generally the real time, latency or flow of the
treatments, but still the cost of dedicated architecture or the cost of the implemented system.
Whatever the target is FPGAs or DSPs.... it remains a difficult semantic phase
of transformation which consists of passing from the abstract type of the variables handled by
our algorithm (real variables, complex, integer, etc…) to a logical type, such as, the bits
vector admissible for RTL system (register logic transfer).
Today, the maitrise of new submicronic technologies, which allows high density
integration of tens of millions of transistors on the same mono-chip, the appearance of a new
stage of design based on the behavioral synthesis which can be assimilated to the design of
source code on the last generations of DSP and FPGA. It consists, starting from the behavioral
specification of an algorithm, in the generation of internal representation (Logic elements) at
the level of Registers Transfer Logic (RTL). The generated code depends on the compilation
techniques of (lexical analysis and syntactic in VHDL, constant propagation etc…) whereas
that the transformations and conversions lie on the methods like scheduling and the
assignment or task of components in order to satisfy the constraints of this application, in
particular, real time processing.
The programming and synthesis software of the FPGA offer convivial tools for an
easy and effective implementation. For an efficient implementation, the use of LUTs
eliminate the necessity of slow classical multiplier, in order to decrease the execution time, i.e
to increase the performance in terms of frequency (sampling throughput). Reducing the LUTs
size, heavy consumer of logic elements, is an efficient solution to remediate at this type of
problem. This leads to optimize the use of hardware resources
KEY WORD: wavelet transform, DWT, denoising, Daubechies filter, Filter banc,
VHDL, FPGA, modelsim, Altera, DE2, Quartus, Matlab.